This paper investigates the task-driven exploration of unknown environments with mobile sensors communicating compressed measurements. The sensors explore the area and transmit their compressed data to another robot, assisting it in reaching a goal location. We propose a novel communication framework and a tractable multi-agent exploration algorithm to select the sensors' actions. The algorithm uses a task-driven measure of uncertainty, resulting from map compression, as a reward function. We validate the efficacy of our algorithm through numerical simulations conducted on a realistic map and compare it with two alternative approaches. The results indicate that the proposed algorithm effectively decreases the time required for the robot to reach its target without causing excessive load on the communication network.